Machine learning has already captured the industry’s attention and driven rapid changes in ad technology, which is the least it could do given the amount of hype it has received. What's even more fascinating, though, is that the pace of the ML revolution is only increasing, and the real change has barely begun. Smart use of ML is now a differentiator and competitive advantage, but it is about to become an absolute requirement to remaining relevant in ad tech.
While there continues to be breakthroughs in core ML research, it is not the academic vanguard that is driving rapid change in our industry, but rather the broadening base of knowledge among nonspecialist engineers. Just a few years ago machine learning was largely restricted to a small group of experts -- a handful of Ph.D.s from a handful of top universities. The ML bottleneck for most ad tech companies was not technology but the recruiting and retention of this rare talent. Since then, simpler developer tools and frameworks have emerged, making it easier for non-specialists to build ML solutions. Along with that, a new wave of learning resources -- online courses, training programs, boot camps, etc. -- have cropped up, all targeted toward general enterprise developers.
Yet even as the tools get simpler, and the courses get better, becoming effective at solving real-world problems with machine learning still doesn’t happen overnight. I know because I have been doing it myself. These days I am a VP of Product Management. I haven’t been a professional, hands-on software developer in 15 years. But I wanted to learn ML and so I dove into the deep end, head first. I needed to brush up on some math I forgot or never took in school. (Tip for a wild and crazy weekend: MIT’s online linear algebra course!) In a little while I found I was learning key concepts of ML and soon, I was able to solve small, but real problems. I figured that if I could learn this stuff, then the smart real engineers at my company could learn it better and faster. We launched an internal program for our engineers called “MLSquared” to address this need. This has proven useful for us because nine months later we have dozens of engineers with real ML skills applying them to some of the most pressing problems in the company.
Other companies are making similar efforts to bolster their core engineer knowledge around ML. Additionally, the tools and training options have been getting better thus easing the process. And now that some time has passed and that group of engineers, people who had not specialized in machine learning at school, but have since been picking up the skills, is getting to a critical mass. The bottleneck of talent is loosening fast. This is a breakthrough!
Suddenly, a lack of resources stemming from limited expertise is no longer an issue and the long list of ML-based projects that had been on the back burner will be able to move forward. It won’t just be the few top projects that can be prioritized, based on a small, elite ML team’s capacity, but ML can be applied to all sorts of projects across the board.
Ad technology is data-heavy. Many of its core issues boil down to optimization problems or pattern identification problems, two areas at which machine learning excels. How much is this impression worth? Who might like to buy it? Is it likely to be fraud? What is this creative about? Is it appropriate? How should I package my inventory? What floors should I set? Etc.
The list of challenges is endless. But with a new generation of ML engineers rapidly emerging, the wait for new solutions to appear on the road map won’t be.
Evan Simeone has more than 15 years of experience in enterprise software engineering and product management. He holds two patents, including one for the notification infrastructure for sending device-specific wireless notifications and another for the apparatus and method for exchanging data between two devices. Evan graduated from New York University.